The New AI Computing Stack: A Guide for Tech Leaders to Navigate Shifting Power Dynamics

Key Takeaways

• AI isn’t a feature you bolt onto your existing infrastructure, but something driving an entirely new computing architecture.
• According to the Forrester report, the traditional three-category vendor landscape (software, cloud, services) has expanded to eight distinct provider categories.
• A new five-layer AI computing stack is forming across experience, orchestration, data, intelligence and infrastructure.
• Vendor lock-in is a real and growing risk, and without a deliberate strategy, your vendors will make architectural decisions for you.
• Open, flexible platforms give technology leaders the best chance to move fast without losing control.

The death of the “bolt-on” era

Every vendor has an “AI version” of something now. An AI-powered dashboard here, a generative feature there. For most enterprises, the approach has been to layer AI onto what already exists: the same cloud stack, the same software vendors, the same procurement playbook.

The problem is that this approach is producing what Forrester researchers Ted Schadler and Bill Martorelli call “Frankenstein” architectures in their September 2025 report, “AI Powers a New Computing Ecosystem”—disconnected tools stitched together without a coherent strategy, driving up costs and complexity at the same time. If your AI spend feels unpredictable, you’re probably living this right now.

The deeper shift Forrester identifies is that AI isn’t just changing how software works. It’s dismantling and rebuilding the technology stack itself. Chips, models, agents and new data infrastructure are redrawing the map of who matters, who holds power and what a healthy vendor relationship even looks like.

For technology leaders, this isn’t just a trend to watch, but a transition to plan for.

Hype vs. reality: what’s actually changing

The confusion in the market is real. Vendors are racing to claim AI territory far outside their traditional lanes. NVIDIA now hosts cloud services. Hyperscalers resell model application programming interfaces (APIs). Software vendors build horizontal agent platforms. The lines between hardware, software and services have blurred, and the pace isn’t slowing.

Here’s what’s worth sifting through the chaos to understand.

First, the vendor landscape has grown dramatically. According to the Forrester report, what used to be three categories at the table—software, cloud and services—has become eight: chipmakers, model builders, hyperscalers, software vendors, data platform providers, agentic platform providers, hardware OEMs and systems integrators. Each is competing for power and control.

Second, AI proof-of-concept projects are no longer enough. Enterprises want measurable results, not more pilots. As Forrester notes in its report, the shift from experimentation to production raises the stakes for every infrastructure decision you make.

Third, some of the most critical AI capabilities now come from vendors you may never contract with directly. Model builders like Anthropic, OpenAI and graphics processing unit (GPU) specialists like NVIDIA are becoming enormously influential, but they typically work through partners rather than selling to enterprises directly. 

Without a clear blueprint, you’re effectively letting your vendors shape your long-term architecture for you. That’s the risk of “agentic sprawl,” and it’s exactly what the Forrester report helps you navigate.

Mapping the new terrain: the five-layer AI computing stack

To build a coherent strategy, you need to understand what the new AI stack actually looks like. The Forrester report identifies five distinct layers, each with its own dynamics and power players.

The infrastructure layer

This is the compute, storage and networking that everything runs on. Traditionally a hyperscaler stronghold, it’s now being disrupted by chipmakers like AMD and NVIDIA, hardware OEMs like Dell Technologies and Hewlett Packard Enterprise and sovereign cloud providers. NVIDIA’s CUDA software gives it control far up the stack, making it the anchor of the AI ecosystem, whether enterprises plan for that or not. Open infrastructure platforms are increasingly important here for teams that want flexibility without full NVIDIA dependency.

The intelligence layer

This is where large language models (LLMs) and reasoning engines live. But hyperscaler/model vendor lock-in is real: Amazon is tied to Anthropic, Microsoft to OpenAI and Google to its own models. If you want to switch models later, migration costs can be high. Forrester recommends abstracting your vector embeddings as much as possible to keep your options open.

The data layer

Data grounds your AI agents in your business context, and it’s often the hardest part of the stack to manage well. Enterprises are reluctant to move large data sets around due to latency and cost. Forrester recommends working with cloud-agnostic suppliers and abstraction layers so you’re not forced to consolidate everything in one location. Your AI and data strategy matters as much as your model choices.

The orchestration layer

This is where AI agents and workflows connect to your systems of record. It’s also where the biggest vendor battles are playing out right now, with software providers like Salesforce, SAP and ServiceNow competing to host the new AI apps. Forrester identifies two paths: advanced organizations will build their own orchestration layer to avoid fragmentation; others will accept packaged solutions from established vendors. Either way, agentic AI won’t replace traditional software, but will work alongside it.

The experience layer

This is where the AI stack meets your users: mobile apps, web interfaces, voice, wearables and more. It’s also where hallucinations and irresponsible AI do the most visible damage. Getting this layer right means applying new design approaches, such as intention translation, linguistic design and context engineering, to build experiences that are accurate and genuinely useful.

Why waiting and watching is a strategic risk

The instinct to wait for the dust to settle is understandable. But Forrester’s analysis makes clear that delay has a real cost.

Incentive structures are already shifting. Consumption-based pricing is replacing seat-based models. Time-and-materials service contracts, according to many of the 23 technology providers Forrester interviewed for its report, are effectively dead. Fixed-price and outcome-based engagements are becoming the norm. If you’re still buying AI infrastructure the old way, you’re probably overpaying.

More importantly, every month without an ecosystem strategy is a month your vendors are making architecture decisions by default. NVIDIA’s CUDA software gives it control far up the AI stack, not just at the chip layer. Hyperscalers are cementing model partnerships. Software vendors are building agent platforms that will be very hard to migrate away from later.

In its report, Forrester identifies seven operating principles for technology leaders navigating this environment. They include acknowledging NVIDIA’s power while minimizing unnecessary dependency, committing to model providers while budgeting for migration contingencies, insisting on multiplayer collaboration across vendors and organizing your stakeholders for the speed of AI decision-making.

SUSE’s approach to open source infrastructure is built for this environment. When vendor lock-in is the primary risk, the ability to run workloads across environments—on-premises, in the cloud, at the edge—gives you the leverage to make deliberate choices. That’s the difference between building the ecosystem you want and inheriting the one your vendors prefer.

For teams managing Kubernetes and container orchestration across hybrid environments, that flexibility becomes even more valuable as AI workloads move into production and the cost of being locked in starts to compound.

Building your AI ecosystem with intent

The shift Forrester describes in its report isn’t really about AI features. It’s about who controls the infrastructure underneath them. Model builders, chipmakers and agentic platforms are each racing to become the indispensable layer in your stack, and each has strong incentives to make switching expensive.

The technology leaders who come out ahead won’t necessarily be the ones who moved fastest. They’ll be the ones who built a deliberate, layered architecture that keeps them in control, with the right partners at each layer of the stack.

That starts with understanding the ecosystem you’re actually operating in. Access the Forrester Report to master the five-layer AI computing stack now and reclaim control of your infrastructure.

Frequently asked questions

What does the new AI computing stack look like?

According to the Forrester report, the new AI computing stack has five distinct layers: infrastructure (compute, storage, networks), intelligence (LLMs and reasoning engines), data (data and knowledge platforms), orchestration (software and agent platforms) and experience (apps, agents and interfaces). Each layer has its own power dynamics and vendor relationships to manage.

How do I prevent my AI infrastructure from becoming a disconnected mess?

The key is building a deliberate orchestration layer that sits outside any single vendor’s vertical stack. Forrester recommends using cloud-agnostic data suppliers with abstraction layers, keeping your vector embeddings portable and setting explicit collaboration frameworks between your ecosystem partners.

Should we build our own AI models or use existing ones?

For most enterprises, the priority isn’t building models, but orchestrating them effectively. Forrester recommends committing to a primary model provider for the best support and integration, while keeping a migration contingency budget and abstracting your data layer to avoid being permanently locked in.